Introduction
Timber canopies that adaptively track daylight can enhance occupant comfort and reduce energy loads; yet, conventional parametric models lack real-time environmental responsiveness. Drawing on phototropism—the growth of plants toward light—this study introduces a neural growth algorithm (NGA) that emulates plant-like directional adaptation to steer parametric canopy geometries in timber architecture. By coupling a spiking neural network to environmental sensors, the system continuously refines panel orientations and supports geometries to balance luminous exposure and structural efficiency.
Methods
An NGA was implemented in Grasshopper using Python and the Elefront plugin. Light intensity data from a grid of photosensors served as input spike trains for a multi-layer spiking neural network, the synaptic weights of which were adapted via a spike-timing–dependent plasticity rule calibrated to phototropic response curves of Helianthus annuus. Each network output triggered geometric transformations of an NURBS-based timber canopy mesh, adjusting panel tilt and curvature. Structural performance was evaluated in Karamba3D through modal and deflection analyses under uniform snow and wind loads. Optimization objectives—maximizing the annual daylight factor (ADF) while limiting maximum deflection to L/360—were encoded as reward signals for the NGA. A baseline static canopy and a gradient-based parametric model (Nelder–Mead model) provided comparative benchmarks.
Results
Over a simulated six-month period in Lusaka’s climate data, the phototropic NGA canopy achieved a 28 % increase in average ADF (from 1.9 % to 2.4 %) and a 17 % reduction in peak deflection compared with the static design. Relative to the gradient-based model, the NGA delivered similar daylight gains (+3 % ADF) but reduced material volume by 12 % through adaptive node pruning. Convergence of the NGA occurred within 120 iterations compared to 350 iterations for the Nelder–Mead model. Real-time adaptation trials demonstrated sub-10-second reconfiguration cycles, enabling near-instantaneous responses to shifting solar angles.
Conclusions
Phototropic-inspired neural growth algorithms offer a compelling bioinspired route for self-adaptive timber canopies, seamlessly integrating environmental sensing with parametric form-finding. The NGA not only improves luminous and structural performance but also streamlines material usage and computation time. Future work will explore integration with mechanoresponsive joints and field validation in full-scale prototypes.
